holoscan-install-conda
Install Holoscan SDK v4.3+ via Conda in a CUDA 13 environment. Use for Conda installs; redirect CUDA 12 hosts to container/wheel.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-10
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- holoscan-install-conda
- license
- Apache-2.0
- version
- 1.0.0
- metadata
- { "tags": [ "holoscan", "install", "conda", "cuda" ], "author": "Holoscan Team <holoscan-team@nvidia.com>", "github-url": "https://github.com/nvidia-holoscan/holoscan-sdk" }
- description
- Install Holoscan SDK v4.3+ via Conda in a CUDA 13 environment. Use for Conda installs; redirect CUDA 12 hosts to container/wheel.
02 资源文件 4
- 参考文档BENCHMARK.md3.8 KB
- 参考文档evals/evals.json4.6 KB
- 参考文档skill-card.md3.6 KB
- 其他skill.oms.sig4.5 KB
03 SKILL.md 正文
Holoscan Conda Installation
Purpose
Install the Holoscan SDK (Python runtime and/or C++ dev headers) into a Conda environment on Linux x86_64, using conda-forge + rapidsai with a correctly pinned CUDA metapackage.
Prerequisites
- Linux x86_64 with an NVIDIA GPU and CUDA 13 driver (check
nvidia-smi). conda(Miniforge preferred). Step 1 installs it if missing.- Network access to conda-forge, rapidsai, and
docs.nvidia.com.
Limitations
- CUDA 13 only (since v4.3.0 — earlier releases were CUDA 12). If the user has a CUDA 12 driver, redirect to
/holoscan-install-containeror/holoscan-install-wheelinstead. - Linux x86_64 only — no aarch64/iGPU support on conda-forge.
ulimit -s 32768is recommended in every shell that runs Holoscan — without it, some apps may segfault.
Step 0: Consult the Official Install Instructions
Always fetch the current Conda section of https://docs.nvidia.com/holoscan/sdk-user-guide/sdk_installation.html before installing — package names, channel selection, and the runtime/dev split can change between releases. Specifically extract:
- The exact runtime package name (e.g.
holoscanfor Python bindings). - The C++ dev package name and whether the user needs it. As of v4.1.0,
libholoscan-devis a separate package containing headers and CMake config — install it whenever the user wants to develop C++ apps. Without it,find_package(holoscan)fails and there are no headers to#include. - Supported Python versions for the current release (3.10–3.13 for v4.3).
- The current
cuda-versionpin (v4.3 →13).
rmm and ucxx are distributed via the rapidsai channel; holoscan, libholoscan, and libholoscan-dev come from conda-forge.
If the doc disagrees with anything below, the doc wins — update the install commands accordingly and tell the user.
Step 1: Prerequisites Check
conda --version 2>&1
nvidia-smi 2>&1 | head -5
If conda is not found, install Miniforge silently (preferred over Miniconda for conda-forge):
wget -q https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-x86_64.sh -O /tmp/Miniforge3.sh
bash /tmp/Miniforge3.sh -b -p ~/miniforge3
source ~/miniforge3/etc/profile.d/conda.sh
conda --version
The -b flag installs non-interactively without modifying .bashrc. Users must source ~/miniforge3/etc/profile.d/conda.sh in each new shell (or add it to their shell RC file) to make conda available.
Step 2: Create Environment and Install
Package roles
libholoscan— C++ runtime symbols (libholoscan_core.so). Auto-pulled as a dependency.holoscan— Python bindings.libholoscan-dev— C++ headers,libholoscan_core.sosymlink, andholoscan-config.cmakeforfind_package(holoscan).rmm— RAPIDS Memory Manager (rapidsai channel). Undeclared runtime dep ofholoscan;import holoscanfails without it.ucxx— UCX Python bindings (rapidsai channel), needed for distributed/multi-process apps.cuda-version=13— pins the CUDA 13 metapackage so the solver picks compatible CUDA runtime libs.
Create the environment first:
source ~/miniforge3/etc/profile.d/conda.sh # if conda not yet on PATH
conda create -n holoscan python=3.13 -y
conda activate holoscan
Then pick one of the variants below based on the user's goal.
Pick the packages for the user's goal — Python-only needs holoscan, C++ dev needs libholoscan-dev, both works for combined use:
conda install <packages> rmm ucxx cuda-version=13 -c rapidsai -c conda-forge -y
For C++ development, also install the toolchain:
conda install -c conda-forge cxx-compiler cmake ninja -y
Verify Python installs with python3 -c "import holoscan; print(holoscan.__version__)". Verify C++ dev installs with ls "$CONDA_PREFIX/include/holoscan".
Step 3: Run Python Tests
ulimit -s 32768 is recommended — without it, some Holoscan apps may segfault on startup.
video_replayer is a display app that loops forever by default. Always patch its YAML to
stop after 10 frames (count: 10, repeat: false, realtime: false) and to run headless
(headless: true) — headless works with or without a display attached and avoids GUI
failure modes over SSH, so we don't branch on $DISPLAY.
Download scripts and YAML configs, patch the YAML, then run:
source ~/miniforge3/etc/profile.d/conda.sh
conda activate holoscan
ulimit -s 32768
SDK_VER=$(python3 -c "import holoscan; print(holoscan.__version__)")
BASE="https://raw.githubusercontent.com/nvidia-holoscan/holoscan-sdk/v${SDK_VER}/examples"
curl -fsSL "${BASE}/hello_world/python/hello_world.py" -o /tmp/hs_hello_world.py
curl -fsSL "${BASE}/video_replayer/python/video_replayer.py" -o /tmp/hs_video_replayer.py
curl -fsSL "${BASE}/video_replayer/python/video_replayer.yaml" -o /tmp/video_replayer.yaml
# Patch video_replayer.yaml — 10 frames, headless.
python3 -c "
c = open('/tmp/video_replayer.yaml').read()
c = c.replace('count: 0', 'count: 10')
c = c.replace('repeat: true', 'repeat: false')
c = c.replace('realtime: true', 'realtime: false')
c = c.replace(' width: 854', ' headless: true\n width: 854')
open('/tmp/video_replayer.yaml', 'w').write(c)"
# hello_world — no display, no data needed; expected: "Hello World!"
python3 /tmp/hs_hello_world.py
# video_replayer — needs racerx data; expected: frames rendered, "Graph execution finished."
HOLOSCAN_INPUT_PATH=/path/to/holoscan/data python3 /tmp/hs_video_replayer.py
HOLOSCAN_INPUT_PATH must point to the directory containing a racerx/ subdirectory.
If the user has the SDK source repo that is ~/repos/holoscan-sdk/data; otherwise download
with the download_ngc_data script from the Debian or source install tree.
Step 4: Remind the User
They must do the following in each new shell session:
source ~/miniforge3/etc/profile.d/conda.sh # if Miniforge was installed with -b
conda activate holoscan
ulimit -s 32768 # recommended — prevents segfaults in some apps
Consider adding these lines to ~/.bashrc or ~/.zshrc to avoid repeating them.
Then offer next steps:
- Explore C++ and Python examples at
https://github.com/nvidia-holoscan/holoscan-sdk/tree/v<VERSION>/examples - Walk through a specific example:
/explain-example - Start building a custom Holoscan application
Troubleshooting
ImportError: librmm.so: cannot open shared object file.rmmwas not installed. Re-run the Step 2conda installline —rmmis an undeclared runtime dependency ofholoscan.- Solver picks an older
holoscanbuild than expected. Channel order may be wrong. Use-c rapidsai -c conda-forge(rapidsai first) — that's the order in the official install command, and under strict channel priority a conda-forge-first ordering can lock the solver to an olderholoscanbuild. - Segmentation fault on app startup. Set
ulimit -s 32768in the current shell before running any Holoscan app. Not all apps trip this, but the larger stack avoids the failure mode. find_package(holoscan)fails when building C++ apps. Installlibholoscan-dev(headers + CMake config are in a separate package since v4.1.0).conda: command not foundin a new shell. Miniforge was installed with-band did not patch.bashrc. Runsource ~/miniforge3/etc/profile.d/conda.shor add it to your shell RC file.
04 同仓库更多 Skills
vercel-sandbox
Run agent-browser + Chrome inside Vercel Sandbox microVMs for browser automation from any Vercel-deployed app. Use when the user needs browser automation in a Vercel app (Next.js, SvelteKit, Nuxt, Remix, Astro, etc.), wants to run headless Chrome without binary size limits, needs persistent browser sessions across commands, or wants ephemeral isolated browser environments. Triggers include "Vercel Sandbox browser", "microVM Chrome", "agent-browser in sandbox", "browser automation on Vercel", or any task requiring Chrome in a Vercel Sandbox.
slack
Interact with Slack workspaces using browser automation. Use when the user needs to check unread channels, navigate Slack, send messages, extract data, find information, search conversations, or automate any Slack task. Triggers include "check my Slack", "what channels have unreads", "send a message to", "search Slack for", "extract from Slack", "find who said", or any task requiring programmatic Slack interaction.
electron
Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) using agent-browser via Chrome DevTools Protocol. Use when the user needs to interact with an Electron app, automate a desktop app, connect to a running app, control a native app, or test an Electron application. Triggers include "automate Slack app", "control VS Code", "interact with Discord app", "test this Electron app", "connect to desktop app", or any task requiring automation of a native Electron application.
dogfood
Systematically explore and test a web application to find bugs, UX issues, and other problems. Use when asked to "dogfood", "QA", "exploratory test", "find issues", "bug hunt", "test this app/site/platform", or review the quality of a web application. Produces a structured report with full reproduction evidence -- step-by-step screenshots, repro videos, and detailed repro steps for every issue -- so findings can be handed directly to the responsible teams.
agent-browser
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.
shadcn
Manages shadcn components and projects — adding, searching, fixing, debugging, styling, and composing UI, including chat interfaces. Provides project context, component docs, and usage examples. Applies when working with shadcn/ui, component registries, presets, --preset codes, or any project with a components.json file. Also triggers for "shadcn init", "create an app with --preset", "switch to --preset", and for building a design system or applying a DESIGN.md or brand spec to shadcn/ui.